Spectral reflectance fitting based on land-based hyperspectral imaging and semi-empirical kernel-driven model for typical camouflage materials

Abstract

Abstract: The reflectance of an object is a physical quantity that is related to a variety of factors such as wavelength, direction of light source, direction of detection, and weather conditions.If complete spectral information about the target is to be obtained, this can only be done by measuring the spectral reflectance in all angular directions. Obviously, this method of acquiring spectral data has the disadvantages of complex operation, low efficiency and poor timeliness in military applications. The Semi-Empirical kernel-driven model captures the main factors affecting the bidirectional reflective properties of an object and uses physically meaningful kernel parameters to characterise the reflective properties of an object. By measuring these kernel parameters and combining them with a small number of measurements, it is possible to extrapolate and fit the spectral reflectance of the target in all directions, improving the efficiency of information acquisition and processing. Semi-empirical kernel-driven models were initially used to study the composition and structure of vegetation and its spectral reflectance properties with some results. However, whether the Semi-empirical kernel-driven model can be effectively used to study the spectral reflectance properties of military materials has not been verified. This paper first introduces three commonly used semi-empirical kernel-driven models, namely RossThick-LiSparseR(RTLSR), RossThick-LiTransitN (RTLT) and RossThick-Roujean (RTR). Then, the spectral reflectance of four typical military materials was measured using an imaging spectrometer, and the fitting effects of different models were evaluated. Experiments show that the three semi-empirical kernel-driven models have good data fitting ability for different types of military materials.Overall, RTLSR model has the best data fitting ability and the best stability of inversion results.